Upload
vonhu
View
214
Download
2
Embed Size (px)
Citation preview
External Economies and International Trade Redux�
Gene M. Grossman
Princeton University
Esteban Rossi-Hansberg
Princeton University
July 2009
Abstract
We study a world with national external economies of scale at the industry level. In
contrast to the standard treatment with perfect competition and two industries, we assume
Bertrand competition in a continuum of industries. In this setting, many of the �pathologies�
of the standard treatment disappear. There typically exists a unique equilibrium with trade
guided by �natural�comparative advantage. And, when a country has CES preferences and
any �nite elasticity of substitution between goods, gains from trade are assured.
�We thank Robert Barro, Arnaud Costinot, Angus Deaton, Elhanan Helpman, Giovanni Maggi, Marc Melitzand four anonymous referees for helpful comments and discussions and the National Science Foundation (undergrant SES 0451712) and Sloan Foundation for research support. Any opinions, �ndings, and conclusions orrecommendations expressed in this paper are those of the authors and do not necessarily re�ect the views of theNational Science Foundation or any other organization.
1 Introduction
External economies hold a central� albeit somewhat uncomfortable� place in the theory of in-
ternational trade. Since Marshall (1879, 1890) at least, economists have known that increasing
returns can be an independent cause of trade and that the advantages that derive from large
scale production need not be con�ned within the boundaries of a �rm. Marshallian externalities
arise when knowledge and other public inputs associated with a �rm�s output spill over to the
bene�t of other industry participants. After Marshall�s initial explication of the idea, a lengthy
debate ensued on whether an industry with external economies of scale could logically be mod-
elled as perfectly competitive.1 Even after the matter was �nally resolved in the a¢ rmative by
Chipman (1970), trade economists continued to bemoan the �bewildering variety of equilibria
[that provide] a taxonomy rather than a clear set of insights� (Krugman, 1995, p.1251), the
apparent tension between the forces of external economies and other, better-accepted determi-
nants of the trade pattern, and the �paradoxical�implication that trade motivated by the gains
from concentrating production need not bene�t the participating countries.
The modeling of external economies began at a time when game theoretic approaches to
imperfect competition were much less developed than they are today. Consequently, the logical
consistency between external economies and perfect competition was seen as a great advantage
in allowing the application of familiar tools of general equilibrium analysis. Economists were
comfortable in assuming that competitors take prices as given in industries with many small
�rms. By assuming that small producers contribute little to aggregate industry output, they felt
similarly justi�ed in assuming that �rms treat industry scale as given when assessing productivity
and making decisions about their own output. Firms might recognize the link between industry
scale and productivity and nonetheless perceive their own costs to be constant. As Chipman
(1970, p.349) noted, �there is no logical contradiction involved in the notion that economic
agents do not perceive things as they actually are, and while the idea of treating production
functions parametrically may be more subtle and unusual than that of treating market prices in
this way, both ideas are of the same logical order.�
In application, the competitive models of external economies and trade yielded some dis-
comforting results.2 A potential for multiple equilibria, noted already by Ohlin (1933, p.55),
was shown more formally by Matthews (1949-1950), Kemp (1964), and Ethier (1982a). Gra-
ham (1923) argued that scale economies could reverse the trade pattern predicted by �natural�
comparative advantage and Ethier (1982a) proved him correct in a competitive model with ex-
ternal economies. Markusen and Melvin (1982) and Ethier (1982a) observed that country size
1For a recount of the debate and its protoganists, see Chipman (1965).2There are several good surveys of this literature, beginning with Chipman�s (1965) treatment of the earliest
contributions. Subsequent overviews include Helpman (1984), Helpman and Krugman (1985, ch.3), Krugman(1995), and Choi and Yu (2003).
1
might play an independent role in determining the pattern of specialization. Finally, Graham
(1923) advanced the possibility of losses from trade, an idea that has been further addressed by
Matthews (1949-1950), Melvin (1969), Markusen and Melvin (1982), Ethier (1982a), and oth-
ers. These �pathologies�are perhaps responsible for the diminished role that external economies
now play in thinking about trade, even as such externalities are still considered to be empirically
relevant in many contexts.3
The pathological outcomes arise in a canonical model of external economies and trade, one
that has become well ensconced in the literature and that by now breeds little questioning of
its underlying assumptions. Typically, there are two countries and two industries, one with
constant returns to scale and the other with increasing returns to scale (IRS) due to national
spillovers from production. Often, labor is the only productive factor. Firms in both sectors are
price takers and, more importantly, those in the IRS sector take aggregate national output and
hence their own potential productivity as given.
In this familiar setting, suppose that the relationship between productivity and national scale
is such that zero output implies zero productivity. Then the existence of multiple equilibria is
immediate. Firms that take aggregate output as given perceive zero productivity when no
other national producers are active in their country. If output in a country is nil, no producer
perceives a pro�t opportunity from entry, even when an absence of local competitors might
imply a potential for pro�t for a �rm that evaluates correctly the implications of entry at non-
negligible scale. Thus, zero output in any location is self-reinforcing and production of the IRS
good can be concentrated in either country for any set of parameter values.4
What about the pattern of trade? When production of the IRS good is concentrated in a
country, the trade pattern necessarily accords with the ranking of relative productivities when
measured at the equilibrium scales of production. That is, each country specializes according
to observed comparative advantage, where the �observed�productivity of a country that does
not produce the IRS good is taken to be zero. However, the pattern of specialization need
not conform to the ranking of relative productivities when measured at a common scale of
production. The exporters of the IRS good might be less productive than �rms in the other
country would be, were the latter to produce at a comparable aggregate scale. In this sense, the
trade pattern may run counter to the dictates of natural comparative advantage. Notice that
such a trade pattern violates the requirements for global e¢ ciency in circumstances in which
the country with comparative advantage in the IRS industry has su¢ cient resources to satisfy
the �rst-best level of world demand for that good.
3See, for example, Caballero and Lyons (1989, 1990, 1992), Chan, Chen and Cheung (1995), Segoura (1996),Henriksen, Midlefart-Knarvik, and Steen (2001) and Antweiler and Tre�er (2002).
4 If productivity is bounded from below even when national output is zero, multiple trade equilibria will existfor a range of parameter values.
2
In the canonical model, country size can be an independent source of comparative advantage.
Suppose, for example, that productivity in the IRS sector is modest when output is small, but
not equal to zero. If a small country were to specialize in producing this good, the equilibrium
price might be high in light of its meager resource base and a robust level of world demand.
Such a high price might invite entry by producers in the large country. In such circumstances,
there may be no equilibrium with production of the IRS concentrated in the small country. The
large country gains a comparative advantage in the IRS industry by dint of having su¢ cient
resources to produce at large scale.
The canonical model readily yields examples with losses from trade (see, e.g., Helpman and
Krugman, 1985, p.55). The autarky equilibrium features underproduction of the IRS good,
because �rms fail to take into account the spillover bene�t from increasing their output on the
productivity of others. As is well known, trade can bring harm if it exacerbates a pre-existing
market distortion (Bhagwati, 1971). It stands to reason that trade may reduce welfare in a
country that shifts substantial resources from the single industry with increasing returns to one
with constant returns.
In short, the pathologies that emerge from the canonical model rest on two critical assump-
tions. First, �rms contemplate entry only at such small scale that they believe they can have
negligible impact on national output and their own productivity. Second, industries are large in
relation to countries�resource endowments, so that a small country may lack su¢ cient resources
to meet world demands for the IRS good at a reasonable price. In such circumstances, the
demise of activity in the IRS industry due to the opening of trade can spell dire consequences
for a country�s productivity and welfare.
In this paper, we modify these two familiar assumptions and show how this dramatically
narrows the scope for pathological outcomes. We assume in what follows that �rms engage in
Bertrand competition and that they correctly anticipate their productivity when they quote a
price that would yield non-negligible sales. And we assume that the world economy comprises
a large number of (small) industries, so that any industry can be fully accommodated in any
country, no matter what its size.5
It is well known that perfect competition and Bertrand competition predict similar outcomes
in models with homogeneous goods and constant costs. This is true as well� it turns out� in an
autarky equilibrium with external economies of scale. However, the predictions of the alternative
behavioral assumptions diverge in a world with national external economies and international
trade. With Bertrand competition, a �rm may be small when its price is the same as that of
5 In our analysis the scope of the external e¤ects is supposed to be the industry. So it is essential for us toassume that industries are small relative to country sizes. As advocated by Ethier (1982b), one could also think ofexternalities that transcend both industry and country boundaries. However, external economies at the industrylevel are plausible in the presence of industry-speci�c knowledge and other non-rival inputs, and the literaturecited in footnote 3 points to their empirical signi�cance.
3
its (domestic) rivals� as will be the case in the equilibria that we discuss� yet it may recognize
that by announcing a price below that o¤ered by other �rms in the industry, it can grow to
non-negligible scale. We shall see that the di¤erent out-of-equilibrium beliefs under perfect and
Bertrand competition yield very di¤erent conclusions about the possibility of multiple equilibria
and the determinants of the trade pattern.
Moreover, the potential role for country size to in�uence the trade pattern disappears when
industries are small in relation to the size of national economies. In a world with many industries
à la Dornbusch, Fischer and Samuelson (1977), a small country can realize its comparative
advantage in whichever sectors it is strongest by serving world demands in a relatively small
range of industries. We will �nd that country size plays no role in determining a chain of
comparative advantage, although it does a¤ect the dividing line between imports and exports
along that chain. The assumption of a large number of small industries also helps to justify
our assumption that �rms take factor prices as given. A Bertrand competitor may entertain the
prospect of growing large in its industry. But, even if it does so, it will remain small in relation to
the aggregate economy so long as each industry uses a small fraction of the economy�s resources.
In summary, we study a model with a continuum of industries and national industry-level
external economies. We assume that �rms in an industry produce a homogeneous product and
engage in Bertrand (price) competition in an integrated world market. Countries may exhibit
Ricardian technological di¤erences, but they share the same prospects for scale economies, which
are assumed to be a property of the good being produced. We �nd in this setting that the trade
equilibrium typically is unique and that �natural�comparative advantage determines an order-
ing of industries.6 Most surprising, perhaps, are our �ndings about welfare. We consider for
this purpose a country whose residents hold CES preferences over the set of goods, with possible
asymmetries in the weighting of di¤erent goods, and with any �nite elasticity of substitution.
This includes, of course, the Cobb-Douglas case that has been used in the literature to demon-
strate the possibility of losses from trade based on external economies of scale. We �nd, in
contrast, that in spite of the generic ine¢ ciency of the autarky and trade equilibria, gains from
trade are assured.
In the main, our analysis follows the older literature in abstracting from two possible aspects
of reality. First, we neglect transport costs. Second, we assume that, given productivity�
which declines with aggregate output� �rms face constant returns to scale. How robust are our
�ndings to changes in these assumptions? We will show that the addition of transport costs
modi�es the analysis somewhat, but important qualitative features survive. When transport
costs are small relative to the potential gains from the realization of increasing returns to scale,
6 If the countries share identical technologies for some or many goods, the model does not predict which goodsare produced in each country. But factor prices and consumption levels typically are uniquely determined evenin this case.
4
the introduction of such costs alters the dividing line between goods produced in each country
but (unlike in Dornbusch, Fischer and Samuelson) does not create a set of nontraded goods. For
larger transport costs, there will be some industries that admit no equilibrium in pure pricing
strategies. In any such industry, the mixed-strategy equilibrium has either all production of the
good in a well-de�ned location, or else production everywhere and an absence of trade.
As regards �rms�constant returns to scale (given productivity), this is a more critical foun-
dation for our results. Our analysis rests on the assumption that a deviant �rm potentially can
internalize the externalities from production if an industry is wrongly located. This requires
that �rms are able to grow large, even if they do not do so in equilibrium. Some small degree
of decreasing returns at the �rm level would not upset our results. But binding limitations on
�rms�expansion would restore the potential for multiple equilibria and some of these would have
patterns of trade counter to the dictates of natural comparative advantage.
The remainder of the paper is organized as follows. The next section develops our alternative
model with Bertrand competition and a continuum of industries. There, we derive our predic-
tions about the pattern of trade. Section 3 contains a brief review of the e¢ ciency properties
of the model, which sets the stage for the gains-from-trade analysis in the subsequent section.
In Section 5, we introduce trading costs, �rst assuming that they are small in relation to the
strength of scale economies, and then allowing for larger costs. Section 6 concludes.
2 A Model of External Economies in a Continuum of Industries
We study an economy with a continuum of industries, as in Dornbusch, Fischer and Samuelson
(1977; hereinafter, DFS). Goods are indexed by i 2 [0; 1]. For the time being, we assume onlythat preferences in each country are such that a change in the prices of a measure-zero set
of goods has a negligible e¤ect on the demand for all goods other than the ones in this set.7
Demands need not be identical in the two countries, nor need preferences be homothetic.
Goods are produced by a single primary factor, labor. Throughout the paper, we take the
home wage as numeraire. Technology exhibits constant or increasing returns due to external
economies at the local industry level. In the home country, ai=Ai(xi) units of labor are required
for a unit of output, where xi is aggregate home production of good i and Ai(�) is non-decreasing,concave, and everywhere has an elasticity smaller than one.8 In the foreign country, the unit
labor requirement is a�i =Ai(x�i ). There are ni potential producers of good i in the home country
and n�i potential producers in the foreign country, where ni � 2 and n�i � 2. The number of
potential producers in each country can be very large, but is assumed to be �nite. Firms in7 In this, we follow Wilson (1980). The Cobb-Douglas preferences invoked by DFS, as well as CES, Stone-Geary,
and symmetric translog all have this property, as do many others.8As in the literature, an elasticity of productivity with respect to output smaller than one is needed to ensure
a non-negative marginal productivity of labor.
5
each industry engage in price (Bertrand) competition in the integrated world market. Each �rm
recognizes that, if it sets a price above that quoted by any other �rm, it will consummate no
sales, if it sets a price that is the lowest among all quoted prices, it will capture the entirety of
world demand, and if it sets a price that is tied for lowest among a group of competitors, it will
share demand with these rivals.
2.1 Autarky Equilibrium
Figure 1 shows the inverse demand curve for good i, labelled DD, for a given level of aggregate
spending and a given level of any price aggregate that may a¤ect the demand for this good.
Firms in the industry take aggregate spending and the relevant price index (if any) as given.
The �gure also depicts the average cost curve, which is labelled CC. Intersection points are our
candidates for industry equilibrium.
0
pi
D
C
D
C
E
xi
pi
xi
~
~
Figure 1: Unique Intersection of DD and CC
Suppose �rms in industry i announce a price above ~pi in Figure 1. Then a deviant �rm could
announce a price a bit lower than that o¤ered by the others. The �rm would capture all industry
sales, and since its price would exceed its average cost at the resulting scale of production, it
would earn a pro�t. Evidently, the price will be bid down at least to ~pi. But a further price
cut by any �rm would spell losses, since the deviant would capture the market but fail to cover
its costs. When the DD curve is everywhere steeper than the CC curve, as in Figure 1, then
the unique intersection point depicts the industry equilibrium. In this equilibrium, price equals
industry average cost and an arbitrary (and possibly large) number of �rms make sales but earn
zero pro�ts.
In Figure 2, the DD and CC curves have multiple intersections at E0, E00, and E. Neither
E0 nor E00 represents an equilibrium, because if all �rms charge the price associated with either
6
of these points, a deviant �rm can announce a lower price such that, at the associated demand
level, price exceeds average cost. The equilibrium in industry i is at the lowest intersection of
DD and CC, so long as DD cuts CC from above.9
0
pi
xi
DC
D
CE
E'
E''''
Figure 2: Multiple Intersections of DD and CC
The general equilibrium requires, of course, that every industry equilibrates and that the
labor market clears. There is no need to develop the notation to express this for the general
case. Instead, we describe a special case that will prove useful later on when we establish the
gains from trade. Suppose utility takes the form
U =
�Z 1
0bic
�i di
�1=�, (1)
where ci is consumption of good i, with � < 1 andR 10 bidi = 1. These preference imply a constant
elasticity of substitution � = 1= (1� �) between every pair of goods. The demand for good i is
ci =b�i p
��i E
P 1��,
where E is aggregate spending and P =hR 10 b
�i p1��i di
i1=(1��)is the exact price index associated
with the CES utility function. With these demands, the condition that DD is steeper than CC
at output xi is equivalent to
��i(xi) < 1,
where �i � A0i(xi)xi=A(xi) is the elasticity of the productivity function with respect to output.9 If no such intersection exists, then the price would tend toward zero and output would tend toward in�nity.
But then the assumption that �rms take wages as given becomes untenable.
7
A su¢ cient condition for existence of an industry equilibrium with �nite xi is ��i(x) < 1 for
all x.10 We will assume that an equilibrium with �nite output exists in every industry, either
because these su¢ cient conditions are satis�ed, or otherwise.
In the autarky equilibrium (denoted again by tildes), aggregate spending ~E equals aggregate
income, L. Then product market equilibrium requires
~xi =b�i ~p
��i L~P 1��
for all i, (2)
while price equals average cost implies
~pi =ai
Ai(~xi)for all i. (3)
This gives a system of equations that jointly determine outputs and prices in all industries.
We note in passing that the autarky equilibrium with Bertrand competition is the same as
would arise under perfect competition, were �rms to take their own productivity as given. Price
in every industry is equal to average cost. Firms earn zero pro�ts. Demand equals supply for
all goods. And the labor market clears.
2.2 Trade Equilibrium
We now open the economy to international trade. Again, competition between the potential
producers in a country drives price for every good i down to average cost. But now it also must
be the case that a potential producer in the �other�country does not wish to shave price further.
If home producers in industry i quote the price wai=A (xi), a foreign �rm could announce a price
a bit lower than that, capture sales of xi, and achieve labor productivity of a�i =Ai(xi). Such a
strategy would be pro�table if its per unit cost, w�a�i =Ai(xi), were less than the quoted price.
Therefore, concentration of industry i in the home country requires
aiAi(xi)
� w�a�iAi(xi)
.
(Note that w = 1, because the home wage is numeraire.) Similarly, if production of good i is
concentrated in the foreign country, it must be that
aiAi(x�i )
� w�a�iAi(x�i )
.
Assuming that each good is produced in only one location (which will be true in equilibrium), the
equilibrium price of a good equals the lesser of the two average costs evaluated at the equilibrium
10Note that Ethier (1982, p.1247) invokes essentially the same assumption as a condition for uniqueness andstabilitiy of equilibrium in his two-industry model.
8
scale of world production; i.e.,
pi = min
�ai
Ai(�xi);w�a�iAi(�xi)
�for all i, (4)
where �xi is world output of good i in the trade equilibrium.
Now order the goods so that �i � ai=a�i is increasing in i and de�ne I by �I = w�. Then
(4) implies that goods with i � I are produced in the home country and goods with i > I
are produced in the foreign country.11 Figure 3� familiar from DFS� depicts this relationship
between w� and I as the curve AA. It depicts as well a second relationship between these
variables implied by the labor-market-clearing condition. The greater is I, the fewer are the
goods produced in the foreign country and the lower must be the foreign wage for the labor
market there to clear. This relationship is labelled BB. The equilibrium marginal good and
relative wage are found at the intersection of the two curves. This determination of I and w�
is, of course, exactly the same as in DFS, as extended by Wilson (1980) to include economies
with more general demands.
0
w*
B
BI
A
A
1
Figure 3: Free Trade Equilibrium
Two observations are in order. First, the pattern of specialization is well determined. The
possibility of multiple locations for a given industry disappears once �rms recognize that, even
if small in equilibrium, each can overcome any coordination failure by becoming large when the
opportunity presents itself. Moreover, there is no danger that a country will lack the resources
to accommodate a particular industry at the scale required to meet world demand. In a world
of many industries, countries can adjust on the extensive margin by hosting greater or fewer
numbers of them, without sacri�cing the bene�ts that come from concentrating an industry in
11For notational convenience, we adopt the convention that the marginal good is produced in the home country,although this is neither determined by the model nor consequential.
9
a single location.
Second, the pattern of specialization conforms to the pattern of comparative advantage, when
the latter is evaluated at a common scale of production. We have assumed in our formulation (as
in the standard approach) that countries share the same scale economies, which are a property
of the industry and not the place of production. Countries di¤er only with respect to a set of
Hicks-neutral (and scale neutral) technology parameters. Then, with no risk of coordination
failure, the country that enjoys the relative technological advantage in an industry will capture
that industry� and its associated scale economies� in the general equilibrium. All that remains
to be determined, as in DFS, is the margin between industries that locate at home and those
that locate abroad. Indeed, the DFS model is a special case of ours with Ai(xi) � 1 for all i.
Here, as there, the dividing line in the chain of comparative advantage is determined by the
relative sizes of the countries, the pattern of world demand, and the pattern of technologies in
the various industries.
Note that country size plays no role in determining relative productivity. Since every individ-
ual industry is small in relation to the size of either economy, any country can produce enough
of any good to meet the pressures of world demand. The relative size of the two countries a¤ects
the margin I between the set of each country�s import goods and its export goods, but it cannot
a¤ect the ordering of goods in those sets.
In short, by incorporating many industries and amending �rms� perceptions about their
own productivity, we have overturned several conclusions from the canonical analysis. Multiple
equilibria are not pervasive and �natural�comparative advantage rules.
A pedagogically interesting special case arises when the countries share identical technologies
and di¤er only in labor supplies. This setting has been used by some authors to argue that trade
induced by external economies will not equalize factor prices in a pair of countries that di¤er
only in size.12 However, this conclusion also rests on the assumption that individual industries
are large in relation to the sizes of the national economies. In our model, when ai = a�i for all
i, if w� 6= 1, then all industries would locate in the low-wage country. This is not consistent, ofcourse, with labor-market clearing in the other country, so factor prices must be equalized. With
equal factor prices and identical technologies, the pattern of specialization is not determined.
What is determined is only the aggregate employment levels in the two countries which, in
equilibrium, matches the exogenous labor supplies. A small country can employ its labor force
in equilibrium by producing and exporting a moderate number of goods produced in relatively
small quantities or a smaller number of goods that require greater numbers of workers to meet
world demand. In any case, di¤erences in country size are accommodated by the number of
goods produced in each country and not by the identities of which goods are produced where.
12See, for example, Markusen and Melvin (1981) and Ethier (1982a).
10
3 E¢ ciency and the First-Best
Both in autarky and in a trade equilibrium, e¢ ciency requires equality between the marginal
rate of substitution and the marginal rate of transformation for every pair of goods. In our
setting, utility-maximizing consumers equate as usual the marginal rate of substitution between
goods to the relative price. But, in general, the marginal rate of transformation is not equal to
the relative price due to the presence of production externalities. Therefore, neither the autarky
equilibrium nor the free-trade equilibrium achieves an e¢ cient allocation of labor.
To see this, note that the marginal product of labor in industry i in the autarky equilibrium
is Ai(~xi)= [ai (1� �i)].13 It follows that the autarky marginal rate of transformation between
good i and good j is
M̂RT ij =
�1� �i(~xi)1� �j(~xj)
�ai=Ai(~xi)
aj=Aj(~xj)=
�1� �i(~xi)1� �j(~xj)
�~pi~pj,
where the second equality follows from (3); i.e. from price equals average cost. Therefore, the
autarky relative outputs of goods i and j can be e¢ cient only if it happens that �i(~xi) = �j(~xj):
Similarly, the free-trade relative outputs can be e¢ cient only if �i(xi) = �j(xj). With relative
prices equal to relative average costs, they will not generally be equal to relative marginal costs.
A set of Pigouvian subsidies could be used to achieve the autarky �rst best. A similar set of
subsidies, together with a set of optimum tari¤s, could do likewise for a country that trades.
It should be clear from the discussion that the autarky equilibrium achieves the �rst best in
the special case in which all industries bear a constant and common degree of scale economies.14
If �i(x) � � for all x and i, the equality between price and average cost in every industry ensuresequality between relative prices and marginal rates of transformation. With trade, the �rst best
still requires a set of optimal tari¤s to improve the terms of trade. But gains from (free) trade
can be established for this special case in the usual way.15
13Let Li be employment in industry i. Then xi = Ai(xi)Li=ai, so dxi = (Ai=ai) dLi + (A0i=ai) dxi or
dxidLi
=Aiai
�1
1� �i
�.
14See Chipman (1970) for an, early, formal demonstration of this point.15For the case of constant and common elasticities �, the free-trade output vector maximizes the value of
domestic output given prices. We can prove gains from trade using the standard, revealed-preference argument:The value of free-trade consumption vector at the prices of the trade equilibrium equals the value of the free-trade output vector at those prices, which exceeds the value of autarky output vector at those prices, which inturn equals the value of the autarky consumption vector at those prices. So, consumers could buy the autarkyconsumption bundle in the trade equilibrium with the income available to them. If they chose not to do so, itmust be that they prefer the trade consumption bundle. Notice that this argument makes no assumptions aboutdemand.
11
4 Gains from Trade
To study the prospects for gains from trade with more general production technologies, we
posit CES preferences for the home country. Domestic preferences are described by the utility
function in (1). We make no assumptions about the weighting of the various goods in utility,
about foreign preferences, or about the productivity functions Ai(xi) (except for the maintained
assumption that Ai(xi) is non-decreasing, concave, and has an elasticity smaller than one). In
particular, we allow for varying degrees of scale economies across industries and non-constant
output elasticities in any or all of them.
4.1 Elastic Demands
We begin with � � 1, which generates demands that are price elastic. Let us suppose that
there are losses from trade; i.e., U < ~U . Since the home wage is numeraire in both the autarky
and trade equilibria, U < ~U requires P > ~P ; i.e., a lower real income with trade than without
implies a higher price index.
0
pi
D
C
D
C
E
D
DC
C
~
~
~
~
~
,xixi~ _
Figure 4: Hypothesized Industry Equilibrium Assuming Losses from Trade
In Figure 4, we show the autarky equilibrium in some industry i. It is found at ~E, the
intersection between ~D ~D and ~C ~C, and has
~xi =b�i ~p
��i L~P 1��
as we have noted before (see Equation 2).
In a world with trade, producers of good i face the world demand
b�i p��i L
P 1��+ c�i (p; w
�L�) ;
12
where c�i (p; w�L�) > 0 is the foreign demand for good i in the trade equilibrium (an arbitrary
function of the vector of prices p and foreign income, w�L�). The hypothesis that P > ~P implies
that P 1�� < ~P 1�� under the assumption that demand is elastic (� � 1). The intuition is thatthe substitution e¤ect of the higher price level outweighs the income e¤ect on the demand for
good i, so the inverse demand curve with trade� DD in the �gure� lies to the right of ~D ~D
under the maintained hypothesis. The average cost curve with trade is depicted by CC and is
given by
min
�ai
A(�xi);w�a�iA(�xi)
�where �xi again is the total world output of good i. This curve coincides with ~C ~C for i � I; i.e.,for goods that are produced by the home country in the trade equilibrium. It lies below ~C ~C
(as depicted in Figure 4) for goods that are produced abroad (i > I); these goods are imported
by the home country only because their average cost is less than it would be with domestic
production. In either case, the intersection of DD and CC lies below and to the right of ~E.
As the �gure makes clear, our hypothesis that P > ~P implies pi < ~pi for an arbitrary good i,
be it one that is exported in equilibrium or one that is imported. Therefore, all prices are lower
in the trade equilibrium than in the autarky equilibrium (relative to the domestic wage). But
this is a contradiction, because the price index cannot rise if each element in the index falls. We
conclude that P < ~P and thus U > ~U ; i.e., the home country gains from trade!
Since P < ~P and � � 1, the inverse demand curve DD may, in fact, be to the left or to
the right of ~D ~D. Apparently, aggregate output in some industries might fall as the result of the
opening of trade. Nonetheless, the impact of these cases cannot be so great as to negate the
overall gains that come from international specialization according to comparative advantage
and the realization of greater scale economies due to the expansion of the market.
4.2 Inelastic Demands
Next consider the case of � < 1; i.e., inelastic demands. Since there are gains from trade as �
approaches one from above16, and the model is continuous in the parameter �, trade losses for
some � would imply the existence of a �̂ such that P (�̂) = ~P (�̂), in the obvious notation. Let us
suppose this to be true, and consider further the autarky and trade equilibria that would arise
when � = �̂.16The limiting case of the CES as � ! 1 is equivalent to Cobb-Douglas preferences. Gains from trade for these
preferences can be established directly. With a constant spending share bi on good i in the home country,
�xiA(�xi)
=biL
min[ai; w�a�i ]+c�i (p; w
�L�)
A(�xi)>biL
ai=
~xiA(~xi)
.
But this implies �xi > ~xi for all i and therefore pi < ~pi for all i. The home worker�s real income increases in termsof every good.
13
The demand function with trade and the price-equals-average cost relationship (4) imply
�xiAi(�xi)
=b�̂i p
1��̂i L
P 1��̂min[ai; a�iw�]+c�i (p; w
�L�)
Ai (�xi)
>b�̂i p
1��̂i L
P 1��̂ai,
where the inequality in the second line follows from the fact that ai � min[ai; a�iw
�] and
c�i (p; w�L�) > 0. Now suppose that pi > ~pi. Then, with �̂ < 1 and P = ~P ,
b�̂i p1��̂i L
P 1��̂ai>b�̂i ~p
1��̂i L
~P 1��̂ai=
~xiAi(~xi)
.
Therefore �xi=Ai(�xi) > ~xi=Ai(~xi). But the function xi=Ai(xi) is increasing in xi by the assumption
that �i(xi) < 1. So the string of inequalities implies �xi > ~xi, which in turn implies pi < ~pi.
This contradicts our supposition that pi > ~pi. We conclude that pi < ~pi for all i, which in turn
implies P < ~P . It follows that there can exist no �̂ for which the price index under autarky is
the same as the price index with trade. Again, the home country gains from trade!
0
pi
D
C
D
C
E
D
DC
C
E
~
~
~
~
~
,xixi~ _
Figure 5: Autarky and Trade Equilibrium for � < 1
Figure 5 illustrates the autarky and free-trade equilibrium in a typical industry i when � < 1.
With P < ~P and � < 1, the DD curve necessarily lies to the right of the ~D ~D curve. The CC
curve is identical to ~C ~C for i � I and strictly below it for i > I. It follows that �xi > ~xi for all i.In this case, trade expands the world output of every good. All prices fall relative to the home
wage, and domestic workers bene�t as a result.
We have established gains from trade for the home country when its residents have CES
preferences, independent of the elasticity of substitution, the pattern of comparative advantage,
the pattern of scale economies, and the details of foreign preferences. This goes some way toward
14
overturning the �paradoxical�welfare implications that derive from the standard analysis, and
the discomfort generated therefrom.
5 Transport Costs
Until now, we have followed the literature on trade with external economies in neglecting trans-
portation costs. In our setting, it might appear that such costs would limit the scope for a
potential producer in a country with natural comparative advantage in an industry to under-
mine the perverse location of that industry in the other country. Might transport costs reintro-
duce the potential for multiple equilibria with di¤erent patterns of trade? We will show that,
when transport costs are small relative to the strength of scale economies, the equilibrium in
every industry is unique and the direction of trade is determined by a combination of natural
comparative advantage, relative market sizes, and transport costs. For larger trading costs,
multiple equilibria can arise, but these do not involve alternative directions of trade. Rather,
the multiplicity exists in industries that lack a pure-strategy equilibrium and it concerns the
probability that a good is produced in a single, well-de�ned location versus the probability that
it is produced everywhere and not traded at all. For every good, trade can �ow in only one
direction.
Before proceeding, we need to specify the market structure that we will consider. Models
with oligopolistic competition and trade are distinguished by whether markets are deemed to be
segmented, so that �rms can announce arbitrarily di¤erent prices in di¤erent geographic loca-
tions, or rather integrated, so that price di¤erences cannot exceed the cost of shipping between
the two markets.17 In our setting, the two market structures yield identical outcomes in the ab-
sence of transport costs, so we did not need to address this issue before now. However, equilibria
for segmented and integrated markets di¤er in the presence of trading costs. We will pursue
here the slightly simpler case of segmented markets, although the interested reader can readily
apply our methods to verify that outcomes with integrated markets are qualitatively similar.
We assume from now on that ti > 1 units of good i must be shipped from any location in
order to deliver one unit of the good to the other country. We also assume that scale economies
are strictly positive (i.e., A0i(�) > 0 for all i). We consider �rst the case in which ti is small forall i, in a sense that will become clear. We then turn to larger trading costs.
5.1 Small Transport Costs
The addition of transport costs introduces the possibility that it will be attractive for a �rm or
�rms to produce only for their local market. Such �rms would enjoy an advantage relative to17See, for example, Venables (1994), who compares outcomes for a Bertrand duopoly with constant production
costs and ad valorem trading costs under the alternatives of segmented and integrated markets.
15
foreign producers inasmuch as they could serve the market without incurring the shipping costs.
However, they would face a disadvantage relative to �rms that serve both markets from their
smaller scale of production. The case of �small�transport costs arises when, for every good, the
former potential advantage from local production is outweighed by the latter cost.
Let us ignore for a moment the possibility that a �rm may target a single market. If all �rms
contemplate selling both locally and abroad, competition among producers in a given location
will bid the pair of prices down to the �rms�cost of serving the respective markets. If home
�rms succeed in capturing both markets, these costs are wai=Ai (xi + x�i ) and waiti=Ai (xi + x�i )
for the home and foreign markets, respectively.18 Then, no deviant �rm in the foreign country
can pro�t by shaving these prices and capturing both markets if�wai
Ai (xi + x�i )� w�a�i tiAi (xi + x�i )
�xi +
�waiti
Ai (xi + x�i )� w�a�iAi (xi + x�i )
�x�i � 0
or ifaia�i� w�
w
tixi + x�i
xi + tix�i. (5)
If, on the other hand, the inequality in (5) runs in the opposite direction, then foreign �rms can
sell at their costs of serving the home and foreign markets, w�a�i ti=Ai (xi + x�i ) and w
�a�iAi (xi + x�i )
respectively, and no deviant home �rm can pro�t by undercutting this pair of prices.
Suppose now that (5) is satis�ed and consider a foreign deviant that seeks to undermine
production in the home country by capturing only its local market. It could do so by announcing
a pair of prices such that its local price were a bit below waiti=Ai (xi + x�i ) and its price for
exporting to home consumers were prohibitively high. This deviation is unpro�table if and only
if �waiti
Ai (xi + x�i )� w�a�iAi (x�i )
�x�i � 0
oraia�i� w�
w
Ai (xi + x�i )
Ai (x�i ) ti. (6)
Analogously, if (5) is violated, then a deviate home �rm will not be able to pro�t by undercutting
foreign producers in (only) the home market if and only if�w�a�i ti
Ai (xi + x�i )� waiAi (xi)
�xi � 0
oraia�i� w�
w
Ai (xi) tiAi (xi + x�i )
. (7)
18Although we continue to take the home wage as numeraire, we will retain w in the expressions in this section,because it improves the clarity of the arguments.
16
Now notice that, when ti is close to one, (5) implies that (6) is satis�ed while the opposite
inequality from (5) implies that (7) is satis�ed.19 In other words, a deviation to serve only one
market can never pay when transport costs are small, because such a deviation requires the
sacri�ce of scale economies.
For ti small, the equilibrium in each industry is unique. In this case, good i is produced
only in the home country if (5) is satis�ed and good i is produced only in the foreign country
if (5) is violated. Competition selects the location for production that minimizes the total cost
of producing and shipping the equilibrium quantities. For ti ' 1, this corresponds to a rankingby natural comparative advantage.20 But for ti somewhat larger, a good that ranks highly in
natural comparative advantage for the home country may nonetheless be produced in the foreign
country, if the aggregate demand for the good is su¢ ciently greater there. What the foreign
producers lose in their comparative disadvantage they potentially can make up with their lesser
shipping costs.
5.2 Large Transport Costs
We have just seen that small trading costs do not threaten trade. As transport costs grow larger,
however, the strategy of pricing to serve only local consumers becomes increasingly attractive.
For ti su¢ ciently large, a deviant can undermine any pure-strategy equilibrium in which output
is concentrated in a single country. Consider, for example, an industry i with trading costs and
other parameters such that (5) is satis�ed but (6) is violated. In such circumstances, good i
cannot be produced only at home, because the implied price to foreign consumers would leave
room for a pro�table deviation by local producers in the foreign country.
We might hypothesize, then, that there exists an equilibrium with production of good i in
both countries and no trade. Indeed, this is correct when ti is su¢ ciently large such that�waiAi (xi)
� waiAi (xi + x�i )
�xi +
�w�a�iAi (x�i )
� waitiAi (xi + x�i )
�x�i � 0 . (8)
The �rst-term in (8) represents the pro�t that a home �rm selling to both markets could earn by
undercutting slightly the local price in a potential equilibrium with no trade. The second term
represents the loss that such a �rm would su¤er by selling abroad despite the high transport
costs. If the sum of these two is negative, an equilibrium with local production in both countries
and no trade in good i would be immune to a pro�table deviation by producers in the home
country that might seek to capture both markets.
But suppose that (6) is violated only slightly; i.e., ai=a�i ' w�Ai (xi + x�i ) =wAi(x�i )ti. Then,19This follows from the fact that (tixi + x�i ) = (xi + tix
�i ) � 1, Ai (xi + x�i ) =A (x�i ) ti � Ai (xi + x�i ) =A (x�i ) > 1,
and Ai (xi) ti=Ai (xi + x�i ) � Ai (xi) =Ai (xi + x�i ) < 1 when ti � 1.
20We use the symbol ' to mean �greater than but approximately equal to�.
17
an equilibrium with dispersed production of good i would be undermined by a deviant in the
home country that shaves prices in both markets. The losses that such a deviant would su¤er
on exports� represented by the second term in (8)� would be quite small. These would be
more than compensated by its strictly positive pro�ts on local sales� the �rst term in (8)� that
would result from its greater productivity compared to its rivals. In short, if (6) is violated only
slightly, then (8) must be violated as well. This situation admits no industry equilibrium in pure
pricing strategies.
In such circumstances, there does exist an equilibrium in mixed strategies. In this equilib-
rium, �rms in one country price at the unit cost they would achieve by serving only their local
market while �rms in the other country randomize among a pair of pricing strategies, one that
yields only local sales and another that ensures sales in both markets. For clarity, we will refer to
these respectively as the �local-pricing strategy�and the �global-pricing strategy.�We proceed
now to characterize the symmetric, mixed-strategy equilibrium; i.e., we describe an equilibrium
in which all �rms in a given country pursue the same (mixed) strategies.
For concreteness, suppose that (5) is satis�ed. Recall that there are ni � 2 home �rms
and n�i � 2 foreign �rms. The latter �rms price so as to compete only for their local market;
i.e., they set a high price for export sales and a local price of w�a�i =Ai (x�i ). The home �rms
randomize. With probability qi, they too compete only for local sales, with a high price for sales
in the foreign country and a local price of wai=Ai (xi). With probability 1 � qi, they contestboth markets, shaving the price w�a�i =Ai (x
�i ) to capture the foreign market, while setting a
price pi > wai=Ai (xi + x�i ) for home sales. The latter strategy entails losses on foreign sales,
but pro�ts on home sales that potentially can compensate.
What values of pi and qi are consistent with equilibrium? First note that pi cannot exceed
wai=Ai (xi). If it did, a deviant �rm could announce a price slightly below pi for home sales,
and a very high price for foreign sales, and thereby earn positive expected pro�ts.21 Also, pi
must be such that each home �rm is indi¤erent between the alternative strategies over which
it randomizes. The local-pricing strategy yields zero pro�ts in all states of the world.22 The
alternative strategy yields pro�ts proportional to
�i (pi) =
�pi �
waiAi (xi + x�i )
�xi +
�w�a�iAi (x�i )
� waitiAi (xi + x�i )
�x�i
where the factor of proportionality re�ects the number of �rms that realize the global-pricing
21With such a strategy, the deviant would earn zero pro�ts whenever any of its rivals quoted the pricewai=Ai (xi). But, with probability (1� qi)ni�1, all would quote the price pi > wai=Ai (xi) for home salesand w�a�i =Ai (x
�i ) for foreign sales. The latter would ensure these �rms of the entire foreign market. The deviant
would capture the bene�ts of large scale without su¤ering any losses on exports, and it would pro�t from its homesales.22 If some rival announces p < wai=Ai (xi), then a �rm quoting wai=Ai (xi) makes no sales and no pro�ts. If no
rival announces this price, then the �rm makes positive local sales, but sells at cost.
18
strategy and thereby share the two markets. Since expected pro�ts from the global-pricing
strategy are proportional to �i (pi), a home �rm will be indi¤erent between the alternative
strategies if and only if�pi �
waiAi (xi + x�i )
�xi +
�w�a�iAi (x�i )
� waitiAi (xi + x�i )
�x�i = 0 . (9)
Note that (9) pins down pi. The second term in (9) is negative, because (6) is violated in
the case under consideration. These are the losses that home �rms su¤er on their foreign sales
as a result of the high transport costs. The �rst term re�ects the positive pro�ts that the home
�rms collectively earn on local sales if they succeed in capturing the world market. Competition
bids the home price down to the point where the home pro�ts just o¤set the foreign losses.
The common mixing probability qi must be such that no home �rm has any incentive to
deviate. Two sorts of potential deviations come into play. First, a home �rm might try to avoid
the losses associated with foreign sales, while still enjoying pro�ts from home sales when some
of its rivals play the global-pricing strategy. Second, a home �rm might deviate to a local price
higher than pi to capitalize on the event that all of its rivals realize the local-pricing strategy.
We address each of these deviations in turn.
Suppose that ni � 1 �rms pursue the mixing described above and consider a deviant thatsets a price just below pi for local sales and a very high price for export sales. Such a deviant
would never sell abroad and thus would never su¤er losses on any export sales. If one or more of
its rivals happen to realize the global-pricing strategy, then these rivals will capture the foreign
market. The deviant home �rm would then capture the home market (thanks to its slightly
lower price), while bene�ting from the external economies generated by its rivals�foreign sales.
In such states, the deviant would earn a positive pro�t. However, if all of its home rivals instead
realize the local-pricing strategy, the deviant will be left with the home market and with a
productivity level that implies losses. The probability of the latter outcome must be su¢ ciently
high so that expected pro�ts from the deviation are non-positive. In particular, we require
�1� qni�1i
��pi �
waiA (xi + x�i )
�xi + q
ni�1i
�pi �
waiAi (xi)
�xi � 0 . (10)
The term in the �rst square brackets in (10) represents the gain from capturing the entirety of
the home market when at least one home rival realizes a price low enough for export sales. The
term in the second square brackets represents the losses that result for the deviant when none of
its home rivals export. Each term is weighted by its respective probability of occurrence. Since
the term in the �rst square bracket is positive and that in the second bracket is negative, there
must be a �nite range of values of qi, including qi = 1, that satis�es (10).
Now consider the deviation that entails a price just below wai=Ai (xi) for home sales and
19
one just below w�a�i =Ai (x�i ) for foreign sales; that is, the deviant prices to ensure itself of export
sales and hopes for an especially large pro�t in the home market by pricing above pi. If no other
home �rm realizes the global-pricing strategy, the deviant will earn positive pro�ts. But if at
least one other home �rms realizes this strategy, the deviant will capture no home sales (due
to its higher price) and yet it will share in the foreign losses. The mixing probabilities for the
others must be such that, overall, this strategy yields the deviant non-positive expected pro�ts,
or that23
qni�1i
�waiA (xi)
� waiA (xi + x�i )
�xi +
1� qniini (1� qi)
�w�a�iAi (x�i )
� waitiAi (xi + x�i )
�xi � 0 . (11)
The left-hand side of (11) is increasing in qi in the range where (6) is violated, which implies
that if (11) is ever satis�ed, it will be so for a range of qi that includes qi = 0:
Any qi that satis�es (10) and (11) can be sustained in a mixed-strategy equilibrium. This
suggests the possibility of multiple equilibria that share certain qualitative features but di¤er in
their mixing probabilities. In all such equilibria, either good i is produced and exported by the
home country or it is not traded at all.
5.3 Implications for the Trade Pattern
We have seen that, when (6) and (7) are satis�ed� as they must be when ti is close to one�
the equilibrium in industry i is unique and always involves trade. If (5) is satis�ed, good i is
produced only in the home country; otherwise, it is only produced in the foreign country. Notice
how this conclusion di¤ers from that in DFS, where transport costs always imply a range of
nontraded goods. In DFS, comparative advantage is the sole motivation for trade. Then, small
transport costs are su¢ cient to choke o¤ trade in industries where comparative cost advantages
are slight. Here, trade takes place not only for reasons of natural comparative advantage, but
also due to scale economies. Small transport costs will not kill trade when external economies
are strictly positive.
When trading costs are larger, industries fall into one of three categories: (i) industries with a
unique pure-strategy equilibrium characterized by production in only one country; (ii) industries
with a unique pure-strategy equilibrium characterized by dispersed production and an absence
of trade; and (iii) industries with a mixed-strategy equilibrium in which either production is
concentrated or the good is not traded.
To see how these possibilities �t together, let us reconsider a good for which (5) holds, i.e.,
a good in which the home country has a su¢ ciently strong natural comparative advantage to
23The �rst term in (11) represents the pro�ts that the deviant earns on home sales when it alone captures exportsales times the probability that this event occurs. The second term represents the deviant�s expected losses onexport sales, considering that it may share the foreign market with zero, one, two, etc. of the other home �rms.
20
make this country the sole candidate for exporting. Recall that the smaller is ai=a�i , the stronger
is the home country�s natural comparative advantage in industry i. If ai=a�i is so small that (6)
also holds, then the industry will be concentrated in the home country, as we have already seen.
No �rm in the foreign country can o¤er a price to local consumers that generates sales and
covers its per-unit costs. If ai=a�i is not very small, but rather large enough that (8) holds, then
production takes place in both countries.24 Let �ai denote the critical value of ai=a�i such that
(8) holds as an equality. Note that �ai may or may not be larger than the left-hand side of (5); if
it is not, then there are no values of ai=a�i for which good i is always nontraded. Finally, for all
values of ai=a�i that satisfy (5) but not (6) or (8), the industry equilibrium must involve mixed
strategies.25 At the lower end of this range, where (6) is just violated, (11) holds only for qi ' 0.At the upper end of the range (if one exists), where ai=a�i / �ai, (10) holds only for qi / 1. Inother words, for parameters such that the condition for an equilibrium with specialization and
trade just fails, the home �rms almost surely choose the global-pricing strategy, whereas for
parameters such that the condition for an equilibrium with dispersed production just fails, the
home �rms almost surely choose the local-pricing strategy.
6 Concluding Remarks
External scale economies need not overturn orthodox presumptions about the pattern and wel-
fare consequences of international trade. By assuming a continuum of industries in which pro-
ducers behave as Bertrand competitors, we have found that the pattern of trade is uniquely
determined and that �natural�productivity advantage governs the chain of comparative advan-
tage. We have also identi�ed a class of preferences, namely CES, for which gains from trade are
assured.
The extensive literature on external economies typically posits a competitive environment
with many small �rms. Researchers have used the small size of producers in a competitive
24For (8) to be satis�ed, we must have
xiAi (xi + x�i )
� xiAi (xi)
+tix
�i
Ai (xi + x�i )> 0
and
aia�i� w�
w
A (xi + x�i )
A (x�i ) ti
2664 tix�i
tix�i � xi�A(xi+x�i )A(xi)
� 1�3775 � �ai .
25The fact thattix
�i
tix�i � xi�A(xi+x�i )A(xi)
� 1� > 1
implies that the right-hand side of (6) is less than �ai: This in turn means that, except when transport costs aresmall, the set of values of ai=a�i at which both (6) and (8) are violated is non-empty.
21
equilibrium to justify an assumption that �rms neglect the impact of their output decisions on
aggregate scale and productivity. We have argued that it is not the equilibrium size of a �rm
that matters for its assessment of productivity, but rather the nature of industry competition
and, in particular, whether a �rm can foresee growing large as a result of its actions. When
industry participants believe that they can grow large, the potential for internalization is enough
to coordinate locational decisions even when externalities remain pervasive in equilibrium.
The literature also typically focuses on an environment with only two industries. Often
the assumption is made for convenience, although it might be appropriate if external scale
economies spread throughout broad sectors. If, instead, external economies are generated by
speci�c activities, a model with many activities may be more appropriate. Then, country size
no longer plays the role that it does in the standard analysis, because a small country can
accommodate all of some activities so long as it hosts relatively few of them.
In this paper, we have considered a static environment in which productivity in an indus-
try and country depends only on current local output. Our results can readily be extended
to settings with dynamic externalities wherein productivity depends also on past experience.
Suppose, for example, that industry productivity is determined by fundamentals and by output
last period. Then, natural comparative advantage (given by fundamentals as well as initial con-
ditions) determines allocations in the �rst period, as in DFS. In the second period, the initial
advantages are strengthened by the specialization pattern of the prior period, and the pattern
of trade gets �locked-in�, as in Krugman (1987). Hence, initial comparative advantage uniquely
determines the subsequent evolution of the economy.26 Moreover, the logic of the present paper
indicates that the outcome will be the same if productivity depends both on current and past
output. Equilibrium in the �rst period would follow natural comparative advantage � as we
have shown� and then the dynamic externalities would lock in this trade pattern forever.
These observations may have more general applicability than our simple Ricardian model.
For example, Grossman and Rossi-Hansberg (2008) study task trade between similar countries.
They assume that �rms perform a continuum of tasks to generate output and that productivity
in each task depends on how frequently it is performed by all �rms in that same location. In that
setting, a �rm can grow large by serving as a potential supplier of a task to others. Although no
outsourcing takes place in equilibrium, the potential for same is enough to coordinate locational
decisions for each task. Moreover, the fact that each task is small means that any task might
be concentrated in a small country.
Another relevant application can be found in urban economics. The existence of cities seems
evidence of agglomeration economies that are, at least partially, external to the �rm. As in
our setup, a site can have small or zero productivity if no �rm locates there, but could be as
26Such an economy could still have several steady states, but the initial conditions will determine a uniquedynamic equilibrium path. See Matsuyama (1991) for a detailed discussion of a similar argument.
22
productive as New York if ten million people reside nearby. Multiple equilibria are pervasive in
models of systems of cities, as sites may or may not host cities depending on whether agents
coordinate to live there. To address this problem, the urban economics literature has invoked the
actions of large �city corporations� (Henderson, 1974) that can subsidize factors to encourage
the internalization of externalities and thereby coordinate location decisions.27 Our theory
provides an alternative mechanism for coordination in naturally productive sites. If �rms engage
in Bertrand competition, each will recognize that by locating alone in a �good� site it can
potentially achieve e¢ cient scale as a �one-company town.�Such a producer might still be small
in relation to the aggregate economy, if the number of towns or cities is large. In equilibrium,
all towns may house numerous �rms, yet the equilibrium allocation of activities across space
will be unique. In short, our conclusions about trade with many industries may apply as well
to agglomeration with many cities.
Our paper demonstrates that the size of industries (or cities) and the nature of the compe-
tition ought to be carefully considered in future modeling of external scale economies.
27Durable housing investments can also serve as a coordinating mechanism in dynamic models, as shown byAnas (1992) and Henderson and Venables (2008).
23
References
[1] Anas, Alex (1992), �On the birth and growth of cities: Laissez-faire and planning com-
pared,�Regional Science and Urban Economics 22, 243-258.
[2] Antweiler, Werner and Tre�er, Daniel (2002), �Increasing Returns and All That: A View
from Trade,�American Economic Review 92, 93-119.
[3] Bhagwati, Jagdish N. (1971), �The Generalized Theory of Distortion and Welfare,�in J.N.
Bhagwati et al., eds., Trade, Balance of Payments and Growth: Papers in International
Economics in Honor of Charles P. Kindleberger (Amsterdam: North-Holland).
[4] Caballero, Ricardo J. and Lyons, Richard K. (1989), �The Role of External Economies in
U.S. Manufacturing,�NBER Working Paper No. 3033.
[5] Caballero, Ricardo J. and Lyons, Richard K. (1990), �Internal versus External Economies
in European Industry,�European Economic Review 34, 805-826.
[6] Caballero, Ricardo J. and Lyons, Richard K. (1992), �The Case for External Economies,�
in A. Cukierman, Z. Hercowitz and L. Leiderman, eds., Political Economy, Growth, and
Business Cycles (Cambridge MA: The MIT Press).
[7] Chan, Vei-Lin , Chen, Been-Lon and Kee-Nam Cheung (1995), �External Economies in
Taiwan�s Manufacturing Industries,�Contemporary Economic Policy 13, 118-130.
[8] Chipman, John S. (1965), �A Survey of the Theory of International Trade: Part 2, The
Neo-Classical Theory,�Econometrica 33, 685-760.
[9] Chipman, John S. (1970), �External Economies of Scale and Competitive Equilibrium,�
Quarterly Journal of Economics 84, 347-363.
[10] Choi, Jai-Young and Yu, Eden S.H. (2002), �External Economies in the International Trade
Theory: A Survey,�Review of International Economics 10, 708-728.
[11] Dornbusch, Rudiger, Fischer, Stanley, and Paul A. Samuelson (1977), �Comparative Advan-
tage, Trade, and Payments in a Ricardian Model with a Continuum of Goods,�American
Economic Review 67, 823-839.
[12] Ethier, Wilfred J. (1982a), �Decreasing Costs in International Trade and Frank Graham�s
Argument for Protection,�Econometrica 50, 1243-1268.
[13] Ethier, Wilfred J. (1982b), �National and International Returns to Scale in the Modern
Theory of International Trade,�American Economic Review 73, 389-405.
24
[14] Graham, Frank D. (1923), �Some Aspects of Protection Further Considered,�Quarterly
Journal of Economics 37, 199-227.
[15] Grossman, Gene M. and Rossi-Hansberg, Esteban (2008), �Task Trade Between Similar
Countries,�Princeton University.
[16] Helpman, Elhanan (1984), �Increasing Returns, Imperfect Markets, and Trade Theory,�in
P.B. Kenen and R.W. Jones, eds., Handbook of International Economics, Vol. 1 (Amster-
dam: North Holland).
[17] Helpman, Elhanan and Krugman, Paul R. (1985), Market Structure and Foreign Trade:
Increasing Returns, Imperfect Competition, and the International Economy (Cambridge
MA: The MIT Press).
[18] Henderson, J. Vernon (1974), �The Sizes and Types of Cities,�American Economic Review
64, 640-656.
[19] Henderson, J. Vernon and Venables, Anthony J. (2008), �The Dynamics of City Formation:
Finance and Governance,�manuscript, Brown University.
[20] Henriksen, Espen R., Midlefart Knarvik, Karen Helene, and Steen, Frode (2001),
�Economies of Scale in European Manufacturing Revisited,�CEPR Discussion Paper No.
2896.
[21] Kemp, Murray C. (1964), The Pure Theory of International Trade and Investment (Engle-
wood Cli¤s: Prentice Hall).
[22] Krugman, Paul R. (1987), �The Narrow Moving Band, the Dutch Disease, and the Com-
petitive Consequences of Mrs. Thatcher : Notes on Trade in the Presence of Dynamic Scale
Economies,�Journal of Development Economics 27, 41-55.
[23] Krugman, Paul R. (1995), �Increasing Returns, Imperfect Competition, and The Positive
Theory of International Trade,�in G.M. Grossman and K. Rogo¤, eds., Handbook of Inter-
national Economics, Vol. 3 (Amsterdam: North Holland).
[24] Matsuyama, Kiminori (1991), �Increasing Returns, Industrialization, and Indeterminacy of
Equilibrium,�The Quarterly Journal of Economics 106, 617-650.
[25] Markusen, James R. and Melvin, James R. (1981), �Trade, Factor Prices, and the Gains
from Trade with Increasing Returns to Scale,�Canadian Journal of Economics 14, 450-469.
[26] Marshall, Alfred (1879), The Pure Theory of Foreign Trade. Reprinted by the London
School of Economics and Political Science, 1930.
25
[27] Marshall, Alfred (1890), Principles of Economics (London: Macmillan and Co.).
[28] Matthews, R.C.O. (1949-1950), �Reciprocal Demand and Increasing Returns,�Review of
Economic Studies 37, 149-158.
[29] Melvin, James R. (1969), �Increasing Returns to scale as a Determinant of Trade,�Canadian
Journal of Economics 3, 389-402.
[30] Ohlin, Bertil (1933), Interregional and International Trade (Cambridge MA: Harvard Uni-
versity Press).
[31] Segoura, Io (1998), �Return to Scale and External Economies: Empirical Evidence from
Greek Two-Digit Manufacturing Industries,�Applied Economic Letters 5, pp.485-490.
[32] Venables, Anthony J. (1994), �Tari¤s and Subsidies with Price Competition and Integrated
Markets: The Mixed-Strategy Equilibria,�Oxford Economic Papers 46, 30-44.
[33] Wilson, Charles A. (1980), �On the General Structure of Ricardian Models with a Contin-
uum of Goods,�Econometrica 48, 1675-1702.
26